Abstract
Specified appliances or middleboxes (MBs) have been explosively used to satisfy a various set of functions in operational modern networks, such as enhancing security (e.g. firewalls), improving performance (e.g. WAN optimized accelerators), providing QoS (e.g. Deep Packet Inspection (DPI)), and meeting the requisite others [3]. Network Function Virtualization (NFV) recently has been proposed to optimize the deployment of multiple network functions through shifting the MB processing from customized MBs to softwarecontrolled inexpensive and commonly used hardware platforms (e.g. Intel standard x86 servers) [4]. However, for some functions (e.g. DPI and Network Deduplication (Dedup), Network Address Translation (NAT)), the commodity shared hardware substrate remain limited performance. For a standard software based Dedup MB (Intel E5645, 2.4GHZ, 6 cores, exclusive mode), we can only achieve 267Mbps throughput in each core at most. Therefore, the resources of dedicated accelerators (e.g. FPGA) are still required to bridge the gap between software-based MB and the commodity hardware.
Original language | English (US) |
---|---|
Title of host publication | Proceedings of the SIGCOMM Chicago 2014 and the Best of the Co-located Workshops |
Editors | Konstantina Papagiannaki |
Publisher | Association for Computing Machinery |
Pages | 353-354 |
Number of pages | 2 |
Volume | 44 |
Edition | 4 |
ISBN (Electronic) | 9781450328364 |
DOIs | |
State | Published - Feb 25 2015 |
Event | ACM SIGCOMM 2014 Conference - Chicago, United States Duration: Aug 17 2014 → Aug 22 2014 |
Conference
Conference | ACM SIGCOMM 2014 Conference |
---|---|
Country/Territory | United States |
City | Chicago |
Period | 8/17/14 → 8/22/14 |
Keywords
- FPGA
- Middlebox
- Network function virtualization
- OpenStack